KR102440442B1 - Vehicle battery life prediction system using vehicle time series data and method thereof - Google Patents

Abstract

The present invention relates to a battery life prediction system and method for an electric vehicle using a battery, such as a hybrid vehicle or an electric vehicle, and more particularly, a time series for the entire cycle from the initial period to the end of the life of the vehicle battery according to the operation of the vehicle Vehicle battery life prediction using vehicle time series data that accurately and quickly predicts the battery life of an electric vehicle by applying new driving data detected in real time to the AI model after creating an AI model that has been trained using data systems and methods.

Description

Vehicle battery life prediction system using vehicle time series data and method thereof

The present invention relates to a battery life prediction system and method for an electric vehicle using a battery, such as a hybrid vehicle or an electric vehicle, and more particularly, a time series for the entire cycle from the initial period to the end of the life of the vehicle battery according to the operation of the vehicle Vehicle battery life prediction using vehicle time series data that accurately and quickly predicts the battery life of an electric vehicle by applying new driving data detected in real time to the AI model after creating an AI model that has been trained using data systems and methods.

In general, engine vehicles such as gasoline and diesel using fossil fuels cause various problems such as environmental pollution due to exhaust gas, global warming due to carbon dioxide, respiratory disease caused by ozone generation, and fuel exhaustion.

To solve these problems, a pure electric vehicle (EV) that uses a battery as a power source and a driving motor as a driving source, and a hybrid electric vehicle (HEV) that runs using an engine and a driving motor as a driving source ), etc., have been developed and commercialized, and the market size of the electric vehicle is expected to grow further due to the global electric vehicle diffusion policy.

In such an electric vehicle, a battery such as a lithium ion battery having advantages of high energy/high power density and long life is used.

However, due to the characteristics of rapid charging and the characteristics of the current profile in which charging and discharging are frequently repeated with various current sizes, deterioration of the lithium-ion battery is inevitable, resulting in changes in the electrochemical characteristics of the battery, that is, reduction in discharge capacity and increase in internal resistance. caused The battery management system (Battery Management System: BMS) is applied, and in particular, accurate estimation and prediction of state-of-health (SOH), which is used as an indicator of deterioration state, is required.

However, deterioration of batteries used in electric vehicles occurs due to the interaction of various influences such as external environment and usage conditions, and it is difficult to predict due to the difficulty and non-linearity of direct measurement of internal electrochemical changes of lithium-ion batteries due to deterioration. There is a problem.

In order to solve this problem, Republic of Korea Patent Registration No. 10-1944313 (2019.02.01. Announcement) [A device for predicting lifespan of a battery] (hereinafter referred to as "Patent Document 1") applies factors input to the life model to extend the lifespan. A life model predictor that predicts and outputs the first predicted value, a real-time life calculator that predicts the lifespan using an electric current-based formula and outputs a second predicted value, and compares the first predicted value with the second predicted value to determine the difference Disclosed is an apparatus for predicting the life of a battery including a comparison predictor for predicting the life of the battery by

However, in Patent Document 1 described above, although it is possible to increase the accuracy by calculating the predicted value double based on the life model and the formula, the first predicted value by the life model and the second predicted value based on the formula are calculated and then compared with each other. Therefore, there is a problem that takes a lot of time.

In addition, since Patent Document 1 is fixedly applied without updating the applied life model, when the life model is incorrectly applied, there is a problem in that an erroneous predicted value can be continuously output.

In addition, in Patent Document 1, it is not possible to know which data the life model is based on, and various mutual influences such as external environment and usage conditions are not reflected therefrom, so that the reliability of the life expectancy value may be lowered.

Therefore, there is a demand for the development of a system capable of predicting the lifespan of a vehicle battery by applying vehicle time series data reflecting various mutual influences such as external environment and usage conditions.

Republic of Korea Patent Registration No. 10-1944313 (2019.02.01. Announcement)

Therefore, it is an object of the present invention to generate an artificial intelligence model that has been trained using time series data for the entire cycle from the beginning of the vehicle battery to the end of life according to the operation of the vehicle, and then use the new driving data detected in real time as the artificial intelligence An object of the present invention is to provide a vehicle battery life prediction system and method using vehicle time series data that accurately and quickly predicts the battery life of an electric vehicle by applying it to a model.

A vehicle battery life prediction system using vehicle time series data according to the present invention for achieving the above object includes: a time series data DB for storing time series data sets for each cycle according to driving of a vehicle; Life prediction model DB that stores the learned life prediction artificial intelligence model; After loading the time series data set from the time series data DB, pre-processing and clustering are performed. Among a plurality of clusters generated by clustering, the amount of data per cycle is large, the number of data in the cluster is large compared to the total number of data, and data for each cycle is included. a learning unit that selects at least one or more clusters, applies the selected cluster to the life prediction AI model to learn, and then stores the learned life prediction AI model in the life prediction model DB; a driving data acquisition unit for acquiring and outputting driving data, which is real-time time series data, from a vehicle; and pre-processing by receiving driving data from the driving data acquisition unit, filtering only driving data corresponding to the cluster selected by the learning unit, and then applying the learned life prediction artificial intelligence model to predict the battery life of the vehicle. It is characterized by including wealth.

The learning unit may include: a data division unit for receiving the time series data and outputting divided time series data that is cut and divided into a size of a unit of a predetermined time interval; a cluster processing unit that converts the divided time series data into divided wavelet data having frequency domain and time domain information by performing wavelet transformation as the preprocessing unit to output the converted wavelet data; a clustering unit that clusters the divided wavelet data output from the cluster processing unit to generate a plurality of clustered clusters; an optimal cluster selector for selecting at least one or more clusters having a large amount of data among the plurality of clusters generated by the clustering unit and having a large number of data in the cluster compared to the total number of data, selecting at least one cluster including data for each cycle, and outputting the selected cluster; a learning processing unit for outputting learning data including the selected cluster data segmentation wavelet data and statistical data extracted from the segmentation wavelet data; and an ensemble learning unit for learning by applying the learning data to the ensemble method of life prediction AI model, and storing the learned life prediction AI model in the life prediction model DB.

The optimal cluster selector divides the selected cluster by a predetermined analogy, divides it into a first part and a second part, outputs the first part to a learning processing unit, and the learning unit, the second part divided by the optimal cluster selector receives the divided wavelet data of , and applies it to the life prediction artificial intelligence model learned in the ensemble method to calculate a life expectancy value, compares the calculated life prediction value with the predicted life expectancy value known in advance for the wavelet data of the second part, and matches After determining whether or not the match rate for the entire wavelet data according to the match determination is equal to or greater than a preset reference match rate, an authentication failure signal is selected as the optimal cluster if the match rate is less than or equal to a preset standard match rate Further comprising a prediction verification unit outputting a negative, wherein the optimal cluster selection unit is characterized in that when the input of the authentication failure signal to change at least some of the selected clusters.

The life prediction unit may include: a driving data division unit configured to receive the driving data and output divided driving data cut by a window having a size of the predetermined time unit; a filtering unit that generates segmented driving wavelet data obtained by performing wavelet transformation on the segmented driving data, then filters and outputs only segmented driving wavelet data corresponding to a cluster selected by the optimal cluster selector and applied to learning; and a prediction unit for outputting a prediction result on the lifespan of a vehicle battery by applying the filtered segmented driving wavelet data to the learned lifespan prediction artificial intelligence model.

The cluster processing unit is characterized in that wavelet transformation is performed by applying wavelet orthogonal (Wavelet bior1.5).

The clustering unit extracts the cluster by performing density-based clustering by applying a density-based spatial clustering of application with noise (DBSCAN).

A vehicle battery life prediction method using vehicle time series data according to the present invention for achieving the above object is: After the learning unit loads the time series data stored in advance from the time series data DB, preprocessing and clustering, and generating by clustering Among the multiple clusters that have been created, the amount of data per cycle is large and the number of data in the cluster is large compared to the total number of data. Then, a learning process of storing the learned life prediction artificial intelligence model in the life prediction model DB; a driving data acquisition process in which the driving data acquisition unit acquires and outputs driving data that is time series data according to a real-time cycle from the vehicle; and a life prediction unit receives driving data from the driving data acquisition unit, performs preprocessing, filters only the preprocessed driving data corresponding to the cluster selected by the learning unit, and applies it to the learned lifespan prediction artificial intelligence model to the vehicle It is characterized in that it includes a life prediction process of predicting the battery life of the.

The learning process may include: a data division step of the learning unit receiving the time series data as the preprocessing through the data division unit, and outputting divided time series data divided into windows having a size of a predetermined time interval unit; a cluster processing step in which the learning unit performs wavelet transformation on the divided time series data as the preprocessing through the cluster processing unit to output divided wavelet data including frequency domain and time domain information; a clustering step in which the learning unit clusters the divided wavelet data output from the cluster processing unit through the clustering unit to generate a plurality of clustered clusters; The learning unit selects at least one or more clusters including the data for each cycle, the amount of data being large among the plurality of clusters generated by the clustering unit and the number of data in the cluster compared to the total number of data, including data for each cycle through the optimal cluster selection unit, an optimal cluster selection step of outputting transformed wavelet data of the cluster; a learning processing step of generating, by the learning unit, statistical data for the segmented wavelet data of the selected cluster through the learning processing unit, and outputting learning data including the statistical data and the segmented wavelet data; and an ensemble learning step of applying the learning data to the lifespan prediction AI model of the ensemble method by the learning unit through the ensemble learning unit and learning, and storing the learned lifespan prediction AI model in the life prediction model DB characterized.

The optimal cluster selection step may include: a selection step in which the learning unit selects an optimal cluster through an optimal cluster selection unit; a separation step of dividing the selected cluster into a first portion and a second portion by a predetermined analogy; an output step of outputting the divided wavelet data of the first part to the learning processing unit and outputting the divided wavelet data of the second part to the prediction verification unit; and a selection cluster change step in which the learning unit changes at least some of the clusters selected when an authentication failure signal is input from the prediction verification unit through the optimal cluster selection unit, wherein the learning process includes the learning unit using the prediction detection unit. The second part of the divided wavelet data divided by the optimal cluster selector is received and applied to the life prediction artificial intelligence model learned in the ensemble method to calculate a life expectancy value, and the calculated life expectancy value and the wavelet data of the second part After determining whether or not the match is determined by comparing the predicted life expectancy value for The method further comprises a prediction verification step of outputting a back-side authentication failure signal to the optimal cluster selector.

The life prediction process may include: a driving data dividing step in which the life prediction unit receives the driving data through the driving data divider and outputs divided driving data cut into a window having a predetermined time unit size; a filtering step of filtering and outputting only the segmented driving wavelet data corresponding to the cluster selected by the optimal cluster selector and applied to learning after the life prediction unit performs wavelet transformation on the segmented driving data through the filtering unit; and a prediction step of outputting a prediction result for the life of the vehicle battery by applying the divided wavelet driving data filtered through the filtering unit to the life prediction artificial intelligence model learned through the prediction unit by the life prediction unit do.

The present invention applies driving data obtained in real time to an artificial intelligence model learned by a time series data set obtained for the entire cycle from the initial application of the vehicle battery to the end of the battery life according to the actual operation of the vehicle. As it predicts the lifespan of the vehicle battery, it has the effect of predicting the lifespan of the vehicle battery more quickly and accurately for various road driving conditions.

In addition, since the present invention generates an artificial intelligence model by selecting and applying an optimal cluster among a plurality of clusters obtained by preprocessing the time series data set, it is possible to more accurately predict the battery life of the vehicle.

In addition, the present invention uses some data of the selected cluster to verify the predicted value predicted in the artificial intelligence model and then, if the verification fails, change at least some of the selected cluster to regenerate the learned artificial intelligence model, so more accurate artificial intelligence By applying the model, it has the effect of more accurately predicting the battery life of a vehicle.

1 is a diagram showing the configuration of a vehicle battery life prediction system using vehicle time series data according to the present invention.
2 is a diagram illustrating the configuration of a learning unit of a vehicle battery life prediction system using vehicle time series data according to the present invention.
3 is a diagram illustrating a configuration of a life prediction unit of a vehicle battery life prediction system using vehicle time series data according to the present invention.
4 is a flowchart illustrating a vehicle battery life prediction method using vehicle time series data according to the present invention.
5 is a flowchart illustrating an artificial intelligence model generation method among the vehicle battery life prediction methods using vehicle time series data according to the present invention.
6 is a flowchart illustrating a lifespan prediction method using an artificial intelligence model among the vehicle battery life prediction methods using vehicle time series data according to the present invention.

Hereinafter, the configuration and operation of a vehicle battery life prediction system using vehicle time series data according to the present invention will be described in detail with reference to the accompanying drawings, and a vehicle battery life prediction method in the system will be described.

1 is a diagram showing the configuration of a vehicle battery life prediction system using vehicle time series data according to the present invention.

The vehicle battery life prediction system using vehicle time series data according to the present invention includes a time series data DB (10), a learning unit (20), a life prediction model DB (30), a driving data acquisition unit (40), and a life prediction unit (50) includes

The time series data DB 10 stores time series data sets for each cycle according to the driving of the vehicle. The cycle-by-cycle time series data set for the entire cycle from the initial time to the end of the life of the vehicle battery according to the driving of the vehicle is the Urban Dynamometer Driving Schedule (UDDS), The Highway Fuel Economy Driving Schedule : HWFET) profile such as battery cell voltage, cell current, real-time mileage, battery power consumption, maximum speed, energy consumption efficiency, and fuel economy data, and preferably any one or more of battery cell voltage and cell current. something to do. The time series data set may have battery remaining life data, ie, cycle information, for each cycle data according to the driving of the vehicle. The remaining battery life may be predicted based on the cycle information.

The learning unit 20 loads the time series data from the time series data DB 10 and performs pre-processing, that is, processing and clustering, and there is a large amount of data among a plurality of clustered clusters and the data for each cycle is evenly distributed. At least one cluster is selected, the selected cluster is applied to the life prediction artificial intelligence model to learn, and then stored in the life prediction model DB 30 .

As for the life prediction AI model, the Ensenble method, which is one of the AI learning methods, is applied, and it is preferable that the XGBoost algorithm is applied among the algorithms for performing the ensemble method. The detailed configuration and operation of the learning unit 20 will be described in detail with reference to FIG. 2 below.

The driving data acquisition unit 40 receives the time series data DB 10 from a vehicle system such as a body control module (BCM) and a battery management system (BMS) of a vehicle when a driving data acquisition event occurs. Acquires and outputs driving data including data corresponding to the data of the time series data set stored in the . The driving data acquisition event may be generated at a predetermined time period when the vehicle is started, or may be generated when a driver clicks a life prediction button provided in the vehicle.

The driving data may be any one or more of cell voltage, cell current, real-time mileage, battery power consumption, maximum speed, energy consumption efficiency, and fuel efficiency obtained from a vehicle battery in real time, and preferably at least one of battery voltage and cell current. something to do.

The life prediction unit 50 extracts driving data corresponding to the selected cluster from among the driving data input from the driving data acquisition unit 40 and applies it to the life prediction artificial intelligence model, the lifespan of the vehicle battery for the driving data Output the prediction result. The life expectancy prediction result may be cycle information for the driving data, or a remaining life corresponding to a cycle of the cycle information for the entire battery life cycle.

2 is a diagram illustrating the configuration of a learning unit of a vehicle battery life prediction system using vehicle time series data according to the present invention.

The learning unit 20 includes a data division unit 21 , a cluster processing unit 22 , a clustering unit 23 , an optimal cluster selection unit 24 , a learning processing unit 25 , and an ensemble learning unit 26 , and , may further include a prediction verification unit 27 according to an embodiment.

The data division unit 21 outputs the divided time series data by cutting the loaded time series data while moving a window having a size of a predetermined time unit by a predetermined time unit. The size of the window may be, for example, 128 bits, and may be configured to divide the time series data while moving in units of 1 bit.

The cluster processing unit 22 generates segmented wavelet data including both time series information and frequency information by wavelet-transforming the segmented time series data. The generated segmented wavelet data is input to the clustering unit 23 . As the wavelet transform method, any one of Wavelet bior1.3, wavelet coiflet 15, and the like may be applied.

The clustering unit 23 generates a plurality of clusters by clustering the divided wavelet data. As the clustering method, a Density-Based Spatial Clustering of Application with Noise (DBSCAN) algorithm method that clusters data according to the degree of data density, that is, density, may be applied, and may be clustered in bits.

The optimal cluster selector 24 selects and outputs an optimal cluster that satisfies a preset condition from among the plurality of clusters generated by the clustering unit 23 . The above condition may be a condition in which data in the cluster must be sufficiently large, and data for each cycle evenly exist within a certain range. That is, the optimal cluster selector 24 selects and outputs at least one or more clusters having a large value of (number of data in cluster)/(total number of data) and including the number of data greater than or equal to the preset number of data for each cycle. There will be. Preferably, the selected population is one.

The optimal cluster selector 24 may be configured to output the cluster of the training set to the learning processing unit 25 by dividing the selected cluster into a training set and a verification set at a certain ratio (eg, 8:2). In addition, the training set may be divided into a training set and a test set at a certain ratio, and only the training set may be output to the learning processing unit 25 .

In addition, when there is only one selected cluster, it may be configured to divide the split wavelet data in the cluster into a training set and a validation set at a certain ratio, and divide the training set into a training set and a test set at a certain ratio.

In another embodiment, the optimal cluster selector 24 may include a high value of (number of data in cluster)/(total number of data) and a preset number of data for each cycle, and a large amount of data for different cycles (that is, different cycles). It may also be configured to select a top certain number of clusters (which are rich in information).

The learning processing unit 25 extracts statistical data for the segmented wavelet data for each cluster, and outputs the extracted statistical data and training data including the segmented wavelet data. The statistical data may be any one of entropy data obtainable by applying segmented wavelet data to the entropy method, median data obtained by applying the median method, and the like.

The ensemble learning unit 26 runs the lifespan prediction AI model, receives the learning data input from the learning processing unit 25 and applies it to the life prediction AI model to perform learning, and the learned lifespan prediction AI The model is stored in the life prediction model DB (30). The life prediction AI model is an ensemble AI model, and may be an XGBoost algorithm or the like.

The prediction verification unit 27 applies the verification set to the life prediction model learned by the ensemble learning unit 26 to generate a life prediction result, and the verification set included in the generated life prediction result and time series data in advance. The lifetime for each cycle, that is, a match is determined by comparing the cycle information, and if the match rate according to the match for the verification set is greater than or equal to a preset reference match rate, the verification is determined to be successful, and if it is below, the verification failure is determined.

When the verification fails, the prediction verification unit 27 outputs the verification failure signal to the optimal cluster selection unit 24 .

When a verification failure signal is input from the prediction verification unit 27 , the optimal cluster selector 24 sequentially changes some or all of the selected clusters. Hereinafter, the term "part or more change" is used to change part or all of it.

Thereafter, the learning data passed through the learning processing unit 25 will be re-learned in the ensemble learning unit 26 to update the lifespan prediction artificial intelligence model.

3 is a diagram illustrating a configuration of a life prediction unit of a vehicle battery life prediction system using vehicle time series data according to the present invention.

The life prediction unit 50 includes a driving data division unit 51 , a filtering unit 52 , and a prediction unit 53 .

The driving data dividing unit 51 receives driving data from the driving data acquiring unit 40, cuts the input driving data into a window having a predetermined time unit and a length of the predetermined time unit, and divides the divided driving data into the filtering unit ( 52) as output.

The filtering unit 52 receives the divided driving data from the driving data dividing unit 51 and performs wavelet transformation to convert the divided driving data into divided wavelet driving data, and among the converted divided driving data, the learning unit ( 20), only the divided driving data corresponding to the selected cluster is filtered and output to the prediction unit 53 .

The prediction unit 53 applies the divided driving data output from the filtering unit 52 to the life prediction artificial intelligence model stored in the life prediction model DB 30 to output a prediction result. The prediction result may be cycle information as described above, or may be residual capacity information of a vehicle's battery. As described above, only the divided driving wavelet data corresponding to the selected cluster is quickly filtered and applied to the lifespan prediction artificial intelligence model, so that the lifespan can be predicted more quickly.

4 is a flowchart illustrating a vehicle battery life prediction method using vehicle time series data according to the present invention.

The learning unit 20 checks whether a learning event occurs (S111). The learning event may occur at a predetermined time period, may occur when time series data is added to the time series data DB 10 , or may occur when a learning command is input by an external administrator.

When a learning event occurs, the learning unit 20 loads and acquires time series data from the time series data DB 10 (S113).

The learning unit 20 applies the loaded time series data to the ensemble artificial intelligence model to learn it, generates the learned ensemble artificial intelligence model, and stores it in the lifespan prediction model DB 30 (S115).

The life prediction unit 50 checks whether a life prediction event occurs ( S117 ). The life prediction event may be generated when driving data is input from the driving data acquisition unit 40 .

When a life prediction event occurs, the life prediction unit 50 receives and acquires driving data from the driving data acquisition unit 40 (S119), and applies the driving data to the life prediction AI model, which is an ensemble AI model, for lifespan predicts and outputs a prediction result (S121).

5 is a flowchart illustrating an artificial intelligence model generation method among the vehicle battery life prediction methods using vehicle time series data according to the present invention. A method of generating an ensemble artificial intelligence model of S115 of FIG. 4 will be described in detail with reference to FIG. 5 .

The learning unit 20 divides the acquired time series data while moving a window having a size of a predetermined time unit (a constant bit unit, eg, 128 bits) in a predetermined time unit (eg, 1 bit unit) to output the divided time series data do (S211). For example, the divided time series data may consist of 128 bits, and when shifted by 1 bit, 127 bits will be the same as the adjacent divided time series data.

The learning unit 20 performs a divided time series data preprocessing process of wavelet-transforming the divided time series data to output divided wavelet data (S213).

When the divided time series data is generated, the learning unit 20 performs clustering by applying density-based, that is, deviscan, to generate a cluster including time series data having a preset density and configured to be adjacent to a predetermined distance (S215).

When a plurality of clusters are generated, the learning unit 20 selects at least one optimal cluster that satisfies a condition in which the number of data is large among the clusters and the data for each sable is evenly distributed in the cluster (S217).

When the optimal cluster is selected, the learning unit 20 performs preprocessing by dividing the data of the selected cluster into a training set, a validation set, and a test set (S219), and generates statistical data for the training set and adds it to the training set A training data preprocessing process for generating one training set is performed (S221).

When the learning data is generated, the learning unit 20 performs learning by applying the learning data to the lifespan prediction artificial intelligence model, and stores the learned lifespan prediction artificial intelligence model in the lifespan prediction model DB 30 (S225) .

When the learned life prediction AI model is stored, the learning unit 20 applies the verification data of the verification set to the learned life prediction AI model to obtain a prediction result, and adds the obtained prediction result and the verification data Pre-stored prediction lifetime, that is, cycle information is compared to determine the degree of agreement, and after calculating the agreement rate for the entire verification data, predictive verification is performed to check whether the calculated agreement rate exceeds or less than a preset reference agreement rate, and (S227), it is determined whether the prediction verification is successful (S229). For example, the reference matching rate may be 80%, and the learning unit 20 may determine that verification is successful when the matching rate for the entire verification data exceeds 80%.

If the verification result is successful, the learning unit 20 ends, and if the verification result fails, it changes the optimal cluster (S231), and then repeats the above-described process after S219.

The learning unit 20 will generate an optimal lifespan prediction artificial intelligence model by changing the optimal cluster until the prediction verification is successful.

6 is a flowchart illustrating a lifespan prediction method using an artificial intelligence model among the vehicle battery life prediction methods using vehicle time series data according to the present invention.

Referring to FIG. 6 , when the driving data is acquired, the life prediction unit 50 divides the acquired driving data into window units to generate divided driving data ( S311 ).

When the divided driving data is generated, the life prediction unit 50 generates divided driving wavelet data by wavelet transforming the divided driving data ( S313 ), and then only the divided driving wavelet data corresponding to the group selected by the learning unit 20 . A filtering process for selecting and outputting is performed (S315).

The life prediction unit 50 applies the filtered divided driving wavelet data to the life prediction artificial intelligence algorithm learned by the learning unit 20 and loaded from the life prediction model DB 30 (S317), and the prediction result is output (S319).

On the other hand, it is common knowledge in the art that the present invention is not limited only to the above-described typical preferred embodiments, but can be improved, changed, replaced, or added in various ways within the scope of the present invention. Those who have will be able to understand it easily. If implementation by such improvement, change, substitution or addition falls within the scope of the appended claims below, the technical idea should also be regarded as belonging to the present invention.

10: Time series data DB 20: Learning part
21: data division unit 22: cluster processing unit
23: clustering unit 24: optimal cluster selection unit
25: learning processing unit 26: ensemble learning unit
27: prediction verification unit 30: life prediction model DB
40: driving data acquisition unit 50: life prediction unit
51: driving data division unit 52: filtering unit
53: prediction unit

Claims (10)

a time series data DB for storing time series data sets per cycle for all cycles from the initial time of the vehicle battery having information on the remaining lifespan for each cycle of the vehicle battery according to the driving of the vehicle to the end of the life of the vehicle battery;
Life prediction model DB that stores the learned life prediction artificial intelligence model;
After loading the time series data set from the time series data DB, preprocessing and clustering are performed, the amount of data per cycle among a plurality of clusters generated by clustering is large, the number of data in the cluster is large compared to the total number of data, and data for each cycle is included a learning unit that selects at least one cluster, applies the selected cluster to the life prediction AI model to learn it, and stores the learned life prediction AI model in the life prediction model DB;
a driving data acquisition unit for acquiring and outputting driving data, which is real-time time series data, from a vehicle; and
A life prediction unit that receives driving data from the driving data acquisition unit and predicts the battery life of the vehicle by pre-processing, filtering only driving data corresponding to the cluster selected by the learning unit, and then applying the learned lifespan prediction artificial intelligence model including,
The learning unit,
a data division unit for receiving the time series data and outputting divided time series data which is cut and divided into a size of a unit of a predetermined time interval;
a cluster processing unit that converts the divided time series data into divided wavelet data having frequency domain and time domain information by performing wavelet transformation as the preprocessing unit to output the converted wavelet data;
a clustering unit that clusters the divided wavelet data output from the cluster processing unit to generate a plurality of clustered clusters;
an optimal cluster selection unit for selecting and outputting at least one cluster having a large amount of data per cycle among the plurality of clusters generated by the clustering unit, a large number of data in clusters compared to the total number of data, and including data for each cycle;
a learning processing unit for outputting learning data including the selected cluster data segmentation wavelet data and statistical data extracted from the segmentation wavelet data; and
Vehicle using time series data, characterized in that it includes an ensemble learning unit for learning by applying the learning data to the life prediction AI model of the ensemble method, and storing the learned life prediction AI model in the life prediction model DB Battery life prediction system.
delete According to claim 1,
The optimal cluster selection unit,
Divide the selected cluster by a certain analogy, divide it into a first part and a second part, and output the first part to a learning processing unit,
The learning unit,
The divided wavelet data of the second part divided by the optimal cluster selector is received and applied to the life prediction artificial intelligence model learned in the ensemble method to calculate a lifetime prediction value, and the calculated lifetime prediction value and the wavelet of the second part After determining whether or not matching is determined by comparing known life expectancy values for data, and determining whether the matching rate for the entire wavelet data of the second part according to the matching determination is equal to or greater than a preset reference coincidence rate, the coincidence rate is set in advance Further comprising a prediction verification unit for outputting an authentication failure signal to the optimal cluster selection unit if it is less than the reference matching rate,
The optimal cluster selection unit,
A vehicle battery life prediction system using vehicle time series data, characterized in that at least some of the selected clusters are changed when the authentication failure signal is input.
According to claim 1,
The life expectancy unit,
a driving data dividing unit for receiving the driving data and outputting divided driving data cut by a window having a size of the predetermined time unit;
a filtering unit that generates segmented driving wavelet data obtained by performing wavelet transformation on the segmented driving data, then filters and outputs only segmented driving wavelet data corresponding to a cluster selected by the optimal cluster selector and applied to learning; and
A vehicle battery life prediction system using vehicle time series data, characterized in that it includes a prediction unit for outputting a prediction result for the vehicle battery life by applying the filtered divided driving wavelet data to the learned lifespan prediction artificial intelligence model.
According to claim 1,
The cluster processing unit,
A vehicle battery life prediction system using vehicle time series data, characterized in that wavelet transformation is performed by applying wavelet orthogonality (Wavelet bior1.5).
According to claim 1,
The clustering unit applies density-based spatial clustering of application with noise (DBSCAN) to extract the cluster by performing density-based clustering.
The learning unit is stored in advance from a time series data DB that stores a time series data set for each cycle for the entire cycle from the initial time of the vehicle battery having information on the remaining lifespan for each cycle of the vehicle battery according to the driving of the vehicle to the end of its life. After loading time series data, preprocess and cluster, and select at least one cluster that has a large amount of data per cycle among a number of clusters generated by clustering, a large number of data in clusters compared to the total number of data, and includes data for each cycle a learning process of applying the selected cluster to the lifespan prediction AI model to learn it, and then storing the learned life prediction AI model in the life prediction model DB;
a driving data acquisition process in which the driving data acquisition unit acquires and outputs driving data that is time series data according to a real-time cycle from the vehicle; and
A life prediction unit receives driving data from the driving data acquisition unit, performs preprocessing, filters only the preprocessed driving data corresponding to the cluster selected by the learning unit, and applies it to the learned life prediction artificial intelligence model of the vehicle. including a life prediction process for estimating battery life;
The learning process is
a data division step of the learning unit receiving the time series data as the preprocessing through the data division unit and outputting divided time series data divided into windows having a size of a predetermined time interval unit;
a cluster processing step in which the learning unit performs wavelet transformation on the divided time series data as the preprocessing through the cluster processing unit to output divided wavelet data including frequency domain and time domain information;
a clustering step in which the learning unit clusters the divided wavelet data output from the cluster processing unit through the clustering unit to generate a plurality of clustered clusters;
The learning unit selects, through the optimal cluster selection unit, at least one or more clusters having a large amount of data for each cycle among a plurality of clusters generated by the clustering unit, a large number of data in a cluster compared to the total number of data, and including data for each cycle, and the selected cluster an optimal cluster selection step of outputting transformed wavelet data of ;
a learning processing step of generating, by the learning unit, statistical data for the segmented wavelet data of the selected cluster through the learning processing unit, and outputting learning data including the statistical data and the segmented wavelet data; and
An ensemble learning step of applying the learning data to the lifespan prediction AI model of the ensemble method by the learning unit through the ensemble learning unit to learn, and storing the learned lifespan prediction AI model in the life prediction model DB A method of predicting vehicle battery life using vehicle time series data.
delete 8. The method of claim 7,
The optimal cluster selection step is
a selection step in which the learning unit selects an optimum cluster through an optimum cluster selection unit;
a separation step of dividing the selected cluster into a first portion and a second portion by a predetermined analogy;
an output step of outputting the divided wavelet data of the first part to the learning processing unit and outputting the divided wavelet data of the second part to the prediction verification unit; and
a selection cluster change step of changing, by the learning unit, at least some of the selected clusters when an authentication failure signal is input from the prediction verification unit through the optimal cluster selection unit,
The learning process is
The learning unit receives the divided wavelet data of the second part divided by the optimal cluster selection unit through the prediction detection unit and applies it to the life prediction artificial intelligence model learned in the ensemble method to calculate a lifetime prediction value, and calculate the calculated lifetime prediction value After determining whether or not the match is determined by comparing the predicted lifetime of the wavelet data of the second part with the wavelet data of the second part, the coincidence rate for the entire wavelet data of the second part according to the match determination is greater than or equal to a preset reference coincidence rate. , A prediction verification step of outputting an authentication failure signal to the optimal cluster selector when the matching rate is less than or equal to a preset reference matching rate.
8. The method of claim 7,
The life expectancy process is,
a driving data dividing step of the life prediction unit receiving the driving data through the driving data dividing unit and outputting divided driving data cut into a window having the predetermined time unit size;
a filtering step of filtering and outputting only the segmented driving wavelet data corresponding to the cluster selected by the optimal cluster selector and applied to learning after the life prediction unit performs wavelet transformation on the segmented driving data through the filtering unit; and
and a prediction step of outputting a prediction result for the life of the vehicle battery by the life prediction unit applying the divided wavelet driving data filtered through the filtering unit to the learned lifespan prediction artificial intelligence model through the prediction unit A method of predicting vehicle battery life using vehicle time series data.

Images